I think you're right. The piledriver cores are big and power hungry (relatively for an embedded application) so Jaguar makes more sense.
At 28nm, I think AMD has shown that they can manufacture a chip up to 352mm (7970) and supply it to the market. Now, I don't think an APU would be that big, but I do think it will end up being close to Pitcrain size ~212mm.
Pitcrain has 20 compute unites (1280 shaders), I think they could sacrifice some of those shaders and use the die area for the cpu core. Maybe 14-16 CU's @ 1GHz and 4 Jaguar cores at 2GHz could fit that 200mm space. That would close to a 2 Tflop SOC, CPU wise way to slow for a PC CPU, but awesome for an embedded system.